Skip to content

ci: rerun on Maven download failures; confirm a solver budget breach before failing - #699

Draft
tstapler wants to merge 3 commits into
mainfrom
fix/ci-flakes
Draft

tstapler wants to merge 3 commits into
mainfrom
fix/ci-flakes

Conversation

@tstapler

@tstapler tstapler commented Nov 20, 2026 •

Copy link
Copy Markdown
Owner

Two CI flakes seen on #696.

  1. Bazel JVM tests died on a Maven registry download (java.io.IOException: Error downloading [https://repo1.maven.org/...] during a rules_kotlin repository fetch). rerun-infra-flakes.yml only recognised Bazelisk/apt/runner signatures, so it called this a real failure and did not rerun. It now also matches that registry-download error (not the 404 "missing artifact" warnings, which Bazel falls through).
  2. SketchEngineBenchmark.solve100Constraints read 58 ms against a 50 ms budget on two runs of feat(sketch): mount the Technical data tab and draw the tool strip icons #696 (observed 7.8 ms normally; the same job passed on ci/quiet-bazel-logs, and the solver is untouched). A shared runner can be starved for a whole best-of-5 window. A breach is now re-measured up to twice more and the lowest figure counts; a real regression stays over the limit in every attempt, so the gate still fails.

Verified: SketchEngineBenchmark (7 tests) and detekt pass locally; scripts/sketch-baseline/check-sketch-baseline.sh passes; actionlint clean. Not verified: the rerun workflow on a real failure.

Update: frame500_should_StayWithinTheRatioToPrebuiltPaths_AndTheCeiling is the second flaky budget (p95 frames of 1.6 to 4.5 ms; x1.54 and x1.83 against a x1.5 limit on #696 and #698, while the solver test read a healthy 18 to 20 ms in the same runs). It gets the same re-measure-before-failing treatment.

@github-actions

github-actions Bot commented Nov 20, 2026 •

Copy link
Copy Markdown
Contributor

JVM Load Benchmark (Desktop)

0 regressed · 1 improved · 4 unchanged · 0 new · 0 removed

jvm/WriteLatency

Benchmark Base Head Change
jankBaselineP95Ms 40 ms 19 ms 🟢 -52.5% ▰▰▰▰▰
4 unchanged

jvm/GraphLoad

Benchmark Base Head Change
phase1TtiMs 0 ms 0 ms ⚪ n/a (base is 0)
phase2Ms 0 ms 0 ms ⚪ n/a (base is 0)
phase3Ms 1 ms 1 ms ⚪ 0.0% ▱▱▱▱▱
totalMs 1 ms 1 ms ⚪ 0.0% ▱▱▱▱▱
Environment
Base Head
branch main HEAD
gitSha 20a23172e 7f178f22f
graphConfig xlarge xlarge
pageCount 230 230

A change counts when it is ≥ 5% and, where the data allows, statistically significant. Judged: 5 by threshold only (point estimates).

@github-actions

github-actions Bot commented Nov 20, 2026 •

Copy link
Copy Markdown
Contributor

JVM Load Benchmark: profiling details

Flamegraphs (this PR) **Allocation** — object allocation pressure (JDBC/SQLite churn)

Alloc flamegraph not available

CPU — method-level hotspots by on-CPU time

CPU flamegraph not available

Top allocation hotspots (this PR) `33.4%` byte[]_[k] `7.1%` java.lang.String_[k] `6.9%` int[]_[k] `5.1%` java.lang.Object[]_[k] `4.8%` java.util.LinkedHashMap$Entry_[k]
Top CPU hotspots (this PR) `98.8%` /usr/lib/x86_64-linux-gnu/libc.so.6 `0.5%` /tmp/sqlite-3.51.3.0-838c454b-196a-408b-9ae0-4e5fa5fae348-libsqlitejdbc.so `0.1%` pthread_cond_signal `0%` SR_handler `0%` prctl

@github-actions

github-actions Bot commented Nov 20, 2026 •

Copy link
Copy Markdown
Contributor

Android Load Benchmark (API 30 x86_64 emulator)

8 regressed · 1 improved · 0 unchanged · 0 new · 0 removed

android/SafIo

Benchmark Base Head Change
safDirectPerFileMs 0 ms 0.03 ms 🔴 n/a (base is 0)
safPerFileMs 0.13 ms 0.23 ms 🔴 +76.9% ▰▰▰▰▰
safOverhead 4 x 7 x 🔴 +75.0% ▰▰▰▰▰

android/WriteLatency

Benchmark Base Head Change
phase3P95Ms 4 ms 18 ms 🔴 +350.0% ▰▰▰▰▰
jankFactor 0.67 x 2 x 🔴 +198.5% ▰▰▰▰▰
baselineP95Ms 6 ms 9 ms 🔴 +50.0% ▰▰▰▰▰
writes 14 writes 23 writes 🟢 +64.3% ▰▰▰▰▰

android/GraphLoad

Benchmark Base Head Change
phase3Ms 3041 ms 5021 ms 🔴 +65.1% ▰▰▰▰▰
phase1Ms 23 ms 36 ms 🔴 +56.5% ▰▰▰▰▰
Environment
Base Head
branch main HEAD
gitSha 20a23172e 7f178f22f
pageCount 530 530

A change counts when it is ≥ 5% and, where the data allows, statistically significant. Judged: 9 by threshold only (point estimates).

@github-actions

github-actions Bot commented Nov 20, 2026 •

Copy link
Copy Markdown
Contributor

CI memory (Gradle diff-coverage job, 16 GB runner)

2 benchmarks (no base to compare with)

ci-memory/diff-coverage

Benchmark Value Range
peak_used [max] 8888 MB —
min_available [min] 7094 MB —
Environment
Head
sha 7f178f22fe3c877248832741a22e035ccbd5638f

Head-only report: add a base to see changes.

@github-actions

Copy link
Copy Markdown
Contributor

CI memory (Bazel Android build, 16 GB runner)

0 regressed · 2 improved · 0 unchanged · 0 new · 0 removed

ci-memory/android-build

Benchmark Base Head Change
min_available [min] 3202 MB 4739 MB 🟢 +48.0% ▰▰▰▰▰
peak_used [max] 12780 MB 11243 MB 🟢 -12.0% ▰▰▱▱▱
Environment
Base Head
sha 20a23172eecf9401021aa5481b3df25f3f83b343 7f178f22fe3c877248832741a22e035ccbd5638f

A change counts when it is ≥ 10% and ≥ 500 absolute and, where the data allows, statistically significant. Judged: 2 by threshold only (point estimates).

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant